Process Repeat Dataset Deliveries

When the business outcome and authorized source scope remain the same, process a recurring delivery in the active dataset-onboarding task. The task retains approved definitions and prior decisions while showing the change introduced by each batch.

Prerequisites and Inputs

Prepare:

• the current active dataset version
• the new batch name, delivery time and source location
• the expected file, worksheet or table inventory
• the data owner, business reviewer, approver and release operator
• a DFS connector and parser that can read the new batch

Compare the New Batch

1. Open the dataset-onboarding task with the active release.
2. Select Check a new delivery.
3. Review the inventory and confirm the expected content is included.
4. Run Profile datasets to inspect fields, volume, missing values and identity evidence.
5. Select Compare with approved version.
6. Review the differences, impact and recommended next action.

Comparison result	Meaning	Next action
Same content	Checked content matches the active version	Retain the active version and record the check
Compatible	Delivery conforms to the approved definitions	Run a new dry run and prepare the release
Review required	A change requires a business decision	Confirm meaning and impact, record the reason, then validate again
Blocking	Structure, identity or relationships have required gaps	Correct the source or approved definition, then inventory and profile again

Review the Changes

Reviewers should check:

• field names, meanings, units and timezone
• volume, missing records and unusual changes
• identifier uniqueness, null meaning and relationships
• MDM identity and protected-field conflicts
• impact on downstream applications

For example, if a field changes from “Temperature” to “Ambient temperature,” the field owner should confirm the measurement location and definition before the team classifies the change.

Dry Run, Publish and Verify

1. Run the dry run and inspect accepted, rejected and quarantined counts.
2. Review the proposed dataset, MDM and fusion effects.
3. Have the approver review the same release evidence.
4. Have the release operator apply the approved version.
5. Verify the active version and downstream application result.

If publication times out or partially fails, refresh the task and inspect each stage before using the recovery action offered for that release.

Complete the Handover

Record the batch name, comparison result, review decision, dry-run summary, active version, downstream check and remaining issues. Deterministic comparison and human review can continue while the model service is temporarily unavailable; AI investigation can resume in the same task after recovery.